The obtained solutions using differential evolution (DE) algorithm (upper bound), where its parameters are tuned by response surface methodology, are compared with exact solutions and computed lower bounds by linear relaxation technique to prove the efficiency of proposed DE algorithm.
International Journal of Industrial Engineering Computations (2013) 127–138 Contents lists available at GrowingScience International Journal of Industrial Engineering Computations homepage: www.GrowingScience.com/ijiec Differential evolution algorithm for multi-commodity and multi-level of service hub covering location problem M EghbaliZarch*, M Abedzadeh and M Setak Department of Industrial Engineering, University of K.N.Toosi, Tehran, Iran ARTICLEINFO Article history: Received 10 August 2012 Received in revised format 14 September 2012 Accepted September 27 2012 Available online 28 September 2012 Keywords: Hub covering location Multi-commodity Multi-level of service Differential evolution algorithm Response surface methodology Lower and upper bounds ABSTRACT The hub location problem involves a network of origins and destinations over which transportation takes place There are many studies associated with finding the location of hub nodes and the allocation of demand nodes to these located hub nodes to transfer the only one kind of commodity under one level of service However, in this study, carrying different commodity types from origin to destination under various levels of services (e.g price, punctuality, reliability or transit time) is studied Quality of services experienced by users such as speed, convenience, comfort and security of transportation facilities and services is considered as the level of service In each system, different kinds of commodities with various levels of services can be transmitted The appropriate level of service that a commodity can be transmitted through is chosen by customer preferences and the specification of the commodity So, a mixed integer programming formulation for single allocation hub covering location problem, which is based on the idea of transferring multi commodity flows under multi levels of service is presented These two are applied concepts, multi-commodity and multi-level of service, which make the model's assumptions closer to the real world problems In addition, a differential evolution algorithm is designed to find near-optimal solutions The obtained solutions using differential evolution (DE) algorithm (upper bound), where its parameters are tuned by response surface methodology, are compared with exact solutions and computed lower bounds by linear relaxation technique to prove the efficiency of proposed DE algorithm © 2012 Growing Science Ltd All rights reserved Introduction Hub location is extensively applied in transportation (Campbell, 2007), telecommunication (Klincewicz, 1998), postal (Çetiner, 2006) and cargo delivery systems (Alumur & Kara, 2008a; & Tan & Kara, 2007) In these systems, to take advantage of the economies of scale, hubs are intermediates for transmitting the flows instead of sending flows directly among all origin-destination pairs Hubs are special facilities, which consolidate and distribute the flows Hub Location Problem (HLP) is introduced by O'Kelly (1987), which can be classified into single and multiple allocation p-hub median, fixed cost, p-hub center and hub covering (Alumur & Kara, 2008a) In hub location problem, the primary objective is to find the best location of hubs and the allocation of * Corresponding author E-mail: meghbali@mail.kntu.ac.ir (M EghbaliZarch) © 2012 Growing Science Ltd All rights reserved doi: 10.5267/j.ijiec.2012.010.001 128 demand nodes to located hubs such that optimize the whole network The primary aim of the p-hub median problem is to minimize the total transportation flow cost between origin-destination pairs and the number of located hubs In HLP with fixed expenditure, the number of hubs is considered as a decision variable The p-hub center is to maximize the minimum cost for (1) every origin-destination pair, (2) movement on every origin to hub, hub to hub and hub to destination link and (3) movement between a hub and an origin/destination In most cases, in hub location problem, every hub pair is connected and there is no direct connection between two demand nodes In fact, not all demand nodes can be allocated to each hub For instance, in air transportation systems low capacity feeder aircraft transport passengers from the low-demand, smaller sized airports to the larger hub airports where passengers are transported to their destinations on bigger capacity aircrafts (Sim, 2007) Hence, capacity feeder limits the number of hub airports where non-hub airports can be assigned to In postal and cargo delivery systems, customer satisfaction occurs when considered commodities are sent from origin to destination in a promised delivery times (Tan & Kara, 2007) Therefore, each demand node cannot be allocated to hubs Campbell (1994) presented hub covering location problem for the first time Kara and Tansel (2003) suggested linear and Wagner (2004) proposed better formulations Ernst et al (2005) introduced cover radius for the first time Hamacher and Meyer (2006) compared different formulations of HCoP and analyzed its feasibility poly-hedron They also proposed a binary search, which was based on the inverse relationship between p-hub center and hub covering problem Tan and Kara (2007) applied the latest arrival hub covering to the cargo delivery sector in Turkey Alumur and Kara (2008b) considered hub covering problem over incomplete network allowing at most four hub stops on a route from any origin to destination, while focusing on cargo application Calik et al (2009) studied single allocation hub covering problem over incomplete hub networks and proposed a tabu-based heuristic algorithm Mohammadi et al (2010) studied single allocation hub covering location problem under a capacity constraint and proposed a hybrid algorithm based on a genetic algorithm and simulated annealing Mohammadi et al (2011b) presented a model for capacitated single allocation hub covering location that instead of using capacity constraints to limit the received flow of each hub, the second objective function was introduced to minimize the service times in the hubs and a weighted-based multi-objective imperialist competitive algorithm was designed to find near-optimal solutions Mohammadi et al (2011a) studied the location model for hub covering problem behaving as M/M/C queues and presented a nonlinear formulation and proposed an improved imperialist competitive algorithm to solve the model Karimi and Bashiri (2011) studied hub covering problem with different coverage types over complete hub networks and proposed two heuristic procedures Zarandi et al (2012) studied the multiple allocation hub set-covering problem considering the backup coverage and mandatory dispersion of hubs Ghodratnama et al (2012) presented a new fuzzy bi-objective model for hub covering problem with specific features such as considering production facilities, time horizons and transportation vehicles Most of the studies in the literature and all of the hub covering location studies in the literature review are dealt with finding the location of hub nodes and the allocation of demand nodes to these located hub nodes to transfer only one kind of commodity under one level of service In reality, each system that hub location can be applied to consist of multi commodities, which transfer commodities under different levels of services Quality of services experienced by users such as speed, convenience, comfort and security of transportation facilities and services is considered as levels of service For instance, in cargo delivery systems, type of commodity can be classified in terms of weight, dimension and nature (like Perishable and non-perishable) that can be transmitted via various kinds of vehicles In postal systems, letters and categorized parcels according to their weight can be posted by various postal services M EghbaliZarch et al / International Journal of Industrial Engineering Computations (2013) 129 The Postal Service divides mail into different services, called classes Each class of mail has different features, postage prices, and presort requirements For most of mailing systems, the content of the material and postage will determine the selected class of mailing In air transportation systems, passengers can choose the service level of their preferred aviation company according to ticket prices, non-delay flights, type of airplanes and reliability of services In telecommunication networks massage or information, in form of voice, data and/or video is transmitted over communication links (e.g copper cables, radio links or satellite links) (Klincewicz 1998) In public transportation, each passenger can be traveled under various modes (e.g walking, driving, bicycling, Commuter Rail / Subway / Bus, taxi, boat) Therefore, in each system with regard to its characteristics, different kinds of commodities with various levels of services can be transmitted The appropriate level of service that a commodity can be transmitted through is chosen by customer preferences and the specification of the commodities The innovation of this research was to consider two concepts of multi-commodity & multi-level of service in hub covering location problem, simultaneously This paper presents a hub covering location problem, which is based on the idea of transferring multi commodity flows under multi levels of service To handle real-size problems, we propose a differential evolution algorithm This paper is organized as follows Section elucidates the mathematical model Section explains the proposed differential evolution algorithm Section illustrates the computational results on Australian Post (AP) data set At the end, conclusions are remarked Mathematical Model The model in this paper is an extension to Ernst and Krishnamoorthy (1999) In the model of Ernst & Krishnamoorthy (1999) only one kind of commodity under one level of service is transmitted in the network, which is not associated with many real-world problems Therefore, we apply two indexes (n & m) to distinct different commodities transmitted under various levels of services with different transportation expenditures Let V be a given node set with v nodes The proposed model locates the pre-determined number of hubs and then assigns non-hub nodes (demand nodes) to locate hubs based on single allocation concept such that each of the origin-hub and hub-destination links meet specified value (cover radius) This model helps to determine the routes for trafficking multi commodity flows under various levels of services from origin to destination through located hubs The objective of proposed model is to minimize the total cost of flows between any origin-destination points and the total cost of establishing hubs The applied parameters and variables of the models are summarized in Table and Table 2, respectively Table Model parameters Parameter Description Number of nodes Number of hubs Cover radius of each hub Discount factor Distance between node i and node k Fixed cost of establishing a hub at node k The unit transportation cost for commodity n under service level m between node i and node j Demand of commodity n under service level m from node i to node j = Total flow of commodity n under service level m originating at node i = Total flow of commodity n under service level m destined for node i 130 Table Model Variables Variable Description equals if node i is allocated to hub k Consistent with the literature, if Xkk=1, it means that node k is a hub The total amount of flow of commodity n under service level m emanating from node i that is routed between hubs k and l By considering previous defined parameters and decision variables, the model is formulated as below: + ( + )+ (1) subject to: = ∀ (3) = ≤ (2) ≤ + = + ∈ 0,1 ∀ , ≥ ∀ , , ∀, (4) ∀i, k (5) ∀, , , (6) (7) , , (8) The objective function minimizes total cost of transferring multi commodity flows under various levels of service between any origin-destination points and the total cost of establishing hubs Eq (2) ensures that every node is assigned to exactly one hub Eq (3) states just p hub can be established Eq (4) ensures that node i cannot be assigned to node k unless there is a hub at node k Eq (5) makes sure that node i can only be allocated to hub k, if distance between i and k is at most the radius Eq (6) is the multi commodity flow balanced under various levels of service equation Eq (7) and (8) state that variable is bigger than zero and is a binary variable Differential Evolution algorithm In this section the details of the DE algorithm is explained and for better understanding of its procedure, an example is implemented It is hard to optimally solve most of the NP-complete problems for real-world instances In this section, a differential evolution algorithm is developed for the presented model The differential evolution (DE) was introduced by Storn and Price(1996) Nowadays the DE has become one of the most frequently used evolutionary algorithms solving the global optimization problems (e.g Rao & Patel, 2012, 2013; Lobato et al., 2012; Vijay Chakaravarthy et al., 2011) Main stages of the DE algorithm are shown in Fig In the literature of DE algorithm, a parent vector from the current generation is called target vector, a mutant vector obtained through the 131 M EghbaliZarch et al / International Journal of Industrial Engineering Computations (2013) differential mutation operation is known as donor vector and finally an offspring formed by recombining the donor with the target vector is called trial vector (Das & Suganthan, 2010) Initialization of Vectors Differencevector based mutation Crossover/Re combination Selection Fig Main stages of the DE algorithm (Das &Suganthan, 2010) 3.1 Representation of the solution Any solution encoding procedure should show the location of hub nodes and the allocation of demand nodes to located hub nodes In this procedure for presenting the given network, some integer numbers are used The solutions are presented as a matrix Each column shows a node in the network, in which its elements value explain the number of the hub and the nodes, which are allocated to them Furthermore, when the value of each element on the entire column is equal to zero, the node is considered as a demand node If the value of the element that is on the same row and the same column is equal to one, that node is considered as a hub and the rest of the elements show the demand nodes allocated to it For example, a sample solution is obtained as follows: = 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 Fig The representing network of the S solution In this solution, node 6, 7, are hubs In addition, node is allocated to hub 6, nodes 1, are allocated to hub and nodes 3, are allocated to hub 3.2 Initial population For creating the initial population, first p numbers of hubs are located randomly and form a hub vector In the allocation step, the remaining nodes are allocated to the located hubs based on their distance from located hubs If a non-hub node can be allocated to more than one hub, it will be allocated to first hub of mentioned hub vector After structuring the whole network, the flow balanced constraint will be checked The above process is applied iteratively to create the entire population 3.3Function evaluation The evaluation is an operation to evaluate how good the network structure of each individual is The evaluation function consists of calculating the objective function value of the network representing by each individual 132 3.4 Mutation Mutation is selected based on the difference of the individual vectors Randomly selected three individuals , and from a group (in generation g) , = + ( , , − , ) (9) Differential Evolution algorithm is based on real-coded, but this model use 0-1 For binary discrete search space, Kennedy and Eberhart (1997) adapted particle swarm optimization (PSO) to search in binary spaces PSO applies a sigmoid transformation to the velocity component of Eq (10) to squash the velocities into a range [0,1], and forces the component values of the locations of particles to be 0’s or 1’s.The equation for updating the positions of Eq (9) is as follows = , , = 1+ (10) , < ( ℎ , ) (11) 3.5 Crossover Crossover is to increase the population diversity and ensures that there is a part of the contribution of the previous generation The crossover operator constructs the offspring , by mixing components of the current individual , (Target matrix) and the , (mutant matrix) generated by mutation There are two types of crossover used in DE, binomial and exponential ones In the proposed algorithm, binomial crossover is applied Binomial crossover replaces the elements of vector , usingthe following rule: Y, = , 0,1 < ℎ , = (12) The feasibility of trial matrix that is computed in this stage should be checked If the trial matrix is not feasible, then it will be necessary to repair it 3.6 Repairing procedure The solution that a trial matrix is shown should be feasible The procedure of repairing the solutions is explained as follow: • If the number of elements with value one on the main diagonal is more than P (more than p hub), then random P elements with value one will be kept and other elements change their values to zero Otherwise, we randomly change their values from elements on the main diagonal with value zero to one until the elements on the main diagonal with value one will be equal to P • The zero value of elements on the main diagonal will force the entire column to get zero values • If the value of element on the main diagonal is equal to one and sum of its entire row is more than one then the elements, which are not located on the main diagonal will change their values to zero • Finally, cover constraint will be checked and if non hub node does not allocated to any hub, then it will be allocated to located hubs according to cover constraint At the end, flow balanced constraint will be checked 3.7 Selection To keep the population size constant over subsequent generations, the next step of the algorithm calls for selection to determine whether the target or the trial vector survives to the next generation, i.e., at G = G + The selection operation is based on comparing , (Trial matrix)and , , , is selected 1333 M EghbaliZarcch et al / Internationall Journal of Industrial Engineering E Computatiions (2013) , , = , ( , ) ≤ ( ) (14) 3.8 Numericaal Example Fo or more dettails about this algoriithm and itts operatorss, an exam mple with nodes andd hubs iss explained Frrom the cu urrent popullation, threee individuaals are sam mpled, rand domly Noow, a scalarr umber F scaales the diffeerences of any a two of these t three individuals i and the scaaled differennce is addedd nu to the third onne to obtain n the donor vector ( , ) In crossoover operattor, the donoor vector exxchanges itss components w with the target vector To keep thhe populatioon size consstant over subsequent generations g s, orithm calls for selecttion to deteermine wheether the tarrget or the trial vectorr thee next step of the algo surrvives to thhe next geneeration (Das & Sugantthan, 2010) This proceedure is app plied on thee mentionedd example as foollow: , = 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 + 0.6 ∗ 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 − 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 = −0.6 6 0.6 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 − 0.6 0 0 0 1 0 0.6 0 0 0 1 0 0 0 0 −0.6 6 0 0 0 1 0.6 0 0 0 − 0.6 0 0 0 1.6 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 Affter random m sampling, three indiviiduals and added a scaled differencee of two off these threee individualss to the third oone are cho osen to obtain the don nor vector A sigmoidd transformation is appplied to thee mponent vaalues to be velocity compponent to sqquash the veelocities intto a range [00, 1] and foorces the com ws in Fig or 1.This proccedure show Fig 33 A sigmoidd transformaation to the veelocity compponent to squuash the veloocities into a range [0, 1] andd force the coomponents values v to be or 134 Ass shown in Fig 4, thee trial vectoor is createed after muutation and crossover stages s Thiss individuaal illu ustrates a ssolution of the probleem and shoould be enteered to thee repairing stage for cchecking itss feaasibility Firrst of all, thhe zero valuue of elements on the main m diagon nal will forcce the entire column too get zero valuues such as column 1,22,3,7 and 8 After thatt, locating P hub is chhecked (It should s be eleements on tthe main diagonal of trrial matrix be equal too one) If thhere is a nodde not alloccated to anyy hu ub (the valuee of the entiire specific row is equaal to zero), then it will be allocateed to one huub accordingg to cover consttraint Checcking cover and flow baalanced connstraint willl be perform med after all of previouss steeps Fig Binoomial crossovver by mixinng componennts of the currrent individuual , (Targget matrix) annd the , (m mutant matrixx) Computattional Resu ult In this sectionn, the propoosed model is implemeented on subbsets of AP dataset AP P dataset haas been usedd baased on a poostal deliverry in Sydneey, Australiia for the fiirst time (Errns & Krishhnamoorthyy, 1996) Too tesst the perforrmance of the t proposed DE algorithm, a com mparison is accomplish hed betweenn the qualityy of algorithm solutions (u upper boundd) and optim mal solutioons, which is i gained byy GAMS233.5 We alsoo M compare Low wer bound co omputed byy linear relax xation (LR)) technique on a system m with Intel(R) Core TM uo T9300 prrocessor (2 50 GHz) with w GB off RAM operrating underr the system m windows Du 4.1 Parameterrs Tuning Th he quality oof an algoriithm is signnificantly afffected by the values of its param meters In this t sectionn, ressponse surfa face method dology (RSM M), an approoach based on DoE tecchniques haas been usedd to tune thee paarameters This T method dology is inttroduced byy Wilson annd Box (19551) for the first time RSM R in thiss 135 M EghbaliZarch et al / International Journal of Industrial Engineering Computations (2013) paper is applied by Design Expert software, which is designed for some experiments to define the promising initial parameter values and leads to improvements in terms of the quality of the solutions 4.2 Computing the Cover Constraint Eq (15) explains Sahraeian and Korani's (2010) approach for calculating the cover radius to apply it in the model By using distance table, the minimum number in each column is obtained (without zero), these values denote vicinity to nodes Also by choosing maximum number among the minimum numbers of columns, cover radius is found = ( ) (15) The mentioned dataset does not have data related to commodity and service level type Hence, Eq (16) shows how to calculate demand of commodity n under service level m from node i to node j and the unit transportation cost for commodity n under service level m between node i and node j is generated randomly close to unit costs of dataset [ = (16) The results of comparison with different values of parameters are shown in Table Table The summary of the performance of the proposed model, GAMS in two circumstances and relative gaps Parameter n=2,m=2 v 15 20 25 40 50 p 5 5 GAMS(MIP) Objective Function Value 633440 846560 1857900 2929400 3774600 5863100 39291000 - DE GAMS(RMIP) metric Upper bound Lower bound Gap1% Gap2% 657863.6816 846564.0177 1927297.4689 2929400 3896139.0315 5895778.945 39450668.4787 61331635.3387 68102880.4356 85400357.3165 633440 826710 1857900 2479300 3460300 4169100 24279000 28761000 38959000 47759000 3.8 3.7 3.2 0.557 0.4 - 3.8 2.4 3.7 18.1 12.5 41.41 62.48 113.2 74.8 78.8 [[ In this section, five instances with a number of nodes, V, ranging from 15 to 50 are considered The optimal values of these instances solved by solver GAMS (Cplex) are given in column under optimal objective function value The column under upper bound give the value of the best solution of DE algorithm and the lower bound column show the value for these instances by relaxing mixed integer programming The next two columns under gap1 and gap2 allow us to compare the bounds and give the percent deviation with respect to the optimal solution and lower bound These two columns consolidate the information of previous columns These two metrics, which are named gap1 and are computed by: 136 1% = 2% = − − × 100 × 100 (17) (18) The obtained results show that since it is no longer possible to optimally solve the instances, then the percent deviation between the upper and lower bound is considered Two mentioned metrics can test the quality of computed bounds and specify how weak they are Tight bounds resulted from upper and lower bound show how good bounds they are By considering the small percent deviation of upper bound and optimal solution in small instances and weak lower bound, it can be concluded that the proposed algorithm has acceptable performance Conclusion In this study, systems carrying different commodity types from origin to destination under various levels of services (e.g price; punctuality, reliability or transit time) have been studied For accommodating the hub covering location problem with the real-world data, it is necessary to apply these two concepts (multi-commodity and multi-level of service) in the models In each system with regard to its characteristics, different kinds of commodity with various levels of service can be transmitted The appropriate level of service that a commodity can be transmitted through is chosen by customer preferences and the specification of the commodity In this paper, a mixed integer programming formulation for the hub covering location problem, which is based on the idea of transferring multi commodity flows under multi levels of service is presented To solve large-sized instances a differential evolution algorithm has been proposed The parameters of the algorithm are tuned by response surface methodology Computational results have compared upper bound (the best solution of DE algorithm) with lower bound and optimal solutions for small instances The percent deviation between upper and lower bounds shows the quality of computed bounds and specify how weak they are By considering the small percent deviation of upper bound and optimal solution in small instances and its weak lower bound, it can be concluded that the upper bound has acceptable quality and the proposed algorithm showed a little gap between the solutions obtained by proposed algorithm and GAMS results References Alumur, SA., & Kara, B.Y (2008a) Network hub location: the state of art.European Journal of Operational Research, 1, 1-21 Alumur, S., & Kara, B.Y.(2008b) A hub covering network design problem for cargo applications in Turkey Journal of the Operational Research Society, 10, 1349-1359 Box, G.E.P., & Wilson, K.(1951) On the experimental attainment of optimum conditions Journal of the Royal Statistical Society Series B (Methodological), 13, 1-45 Campbell, J.F.(1994) Integer programming formulations of discrete hub location problems.European Journal of Operational Research, 2, 387-405 Campbell, J.F (2009) Hub location for time definite transportation Computers & Operations Research, 36, 3107-3116 Calik, H., Alumur, SA., Kara, B.Y & Karasan, O.E.,(2009) A tabu-search based heuristic for the hub covering problem over incomplete hub networks Computers & Operations Research,12, 30883096 M EghbaliZarch et al / International Journal of Industrial Engineering Computations (2013) 137 Çetiner, S., Sepil, C., & Süral, H (2010) Hubbing and routing in postal delivery systems Annals of Operations Research, 181, 109-124 Das, S., & Suganthan, P.N (2010) Differential evolution: A survey of the state-of-the-art Evolutionary Computation, IEEE Transactions, 99, 1-28 Ernst, A., Jiang, H & Krishnamoorthy, M.,(2005) Reformulations and computational results for uncapacitated single and multiple allocation hub covering problems.Unpublished Report, CSIRO Mathematical and Information Science Ernst, A.T., & Krishnamoorthy, M (1996) Efficient algorithms for the uncapacitated single allocation p-hub median problem Location Science,4, 139-154 Ernst, A.T., &Krishnamoorthy, M (1999) Solution algorithms for the capacitated single allocation hub location problem Annals of Operations Research, 86, 141-159 Fazel Zarandi, M., Davari, S.& Haddad Sisakht, S (2012) The Q-coverage multiple allocation hub covering problem with mandatory dispersion Scientia Iranica Ghodratnama, A., Tavakkoli-Moghaddam, R., & Azaron, A (2012) A fuzzy possibilistic bi-objective hub covering problem considering production facilities, time horizons and transporter vehicles The International Journal of Advanced Manufacturing Technology, 1-20 Hamacher, H.W., & Meyer, T (2006) Hub cover and hub center problems Departament of Mathematics, University of Kaiserslautern, Germany Klincewicz, J.G (1998) Hub location in backbone/tributary network design: a review Location Science, 6, 307-335 Kennedy, J., &Eberhart, R.C.(1997) A discrete binary version of the particle swarm algorithm IEEE TRANSACTION Karimi, H., & Bashiri, M (2011) Hub covering location problems with different coverage types Scientia Iranica,18, 1571–1578 Kara, B., & Tansel, B (2003) The single-assignment hub covering problem: Models and linearizations.Journal of the Operational Research Society, 59-64 Lobato, F.S., Gedraite,R & Neiro, S.M.S.,(2012) Solution of flow shop scheduling problems using the differential evolution algorithm.3rd International Conference on Engineering Optimization, Rio de Janeiro, Brazil Mohammadi, M., Jolai, F & Rostami, H (2011a) An M/M/c queue model for hub covering location problem.Mathematical and Computer Modelling, 54, 2623–2638 Mohammadi, M., Tavakkoli-Moghaddam, R & Rostami, H (2011b) A multi-objective imperialist competitive algorithm for a capacitated hub covering location problem International Journal of Industrial Engineering Computations 2, 671-688 Mohammadi, M., Tolui, H & Yousefi, M (2010) Solving a hub covering location problem under capacity constraints by a hybrid algorithm Journal of Applied Operational Research, 2, 109-116 O'kelly, M.E (1987) A quadratic integer program for the location of interacting hub facilities.European Journal of Operational Research, 3, 393-404 Rao, R.V & Patel, V.,(2012) An elitist teaching learning based optimization algorithm for solving complex constrained optimization problems.International Journal of Industrial Engineering Computations, 3, 535-560 Rao, R.V & Patel,V (2013) Comparative performance of an elitist teaching-learning-based optimization algorithm for solving unconstrained optimization problems International Journal of Industrial Engineering Computations Sim, T.K.T (2007) The hub covering flow problem and the stochastic p-hub center problem United states: ProQuest Information and Learning Company Tan, P.Z., Kara, B.Y (2007) A hub covering model for cargo delivery systems Networks, 1, 28-39 Wagner, B.,(2004) Model formulations for hub covering problems Journal of the Operational Research Society, 7, 932-938 Storn, R., &Price, K., (1996) Minimizing the real functions of the ICEC'96 contest by differential evolution IEEE Transaction 138 Sahraeian, R., & Korani, E (2010) The hierarchical hub maximal covering problem with determinate cover radiuses IEEE Transaction Vijay Chakaravarthy, G., Marimuthu, S., & Naveen Sait, A (2011) Performance evaluation of proposed Differential Evolution and Particle Swarm Optimization algorithms for scheduling mmachine flow shops with lot streaming Journal of Intelligent Manufacturing, 1-17 ... every origin to hub, hub to hub and hub to destination link and (3) movement between a hub and an origin/destination In most cases, in hub location problem, every hub pair is connected and there is... between p -hub center and hub covering problem Tan and Kara (2007) applied the latest arrival hub covering to the cargo delivery sector in Turkey Alumur and Kara (2008b) considered hub covering problem. .. levels of services Quality of services experienced by users such as speed, convenience, comfort and security of transportation facilities and services is considered as levels of service For instance,